Translational Knowledge: From Collecting Data to Making

Los Alamos National Labs, Dec 7, 2010

Translational Knowledge: From Collecting Data to Making Decisions in a Smart GridData to Making Decisions in a Smart Grid

Mladen KezunovicDepartment of Electrical and Computer Engineering

Texas A&M UniversityTexas A&M [email protected]

©2010 Mladen Kezunovic, All Rights Reserved

• BackgroundO i l F l L i• Optimal Fault Location

• Intelligent Alarm ProcessingI h tl Ad ti F lt• Inherently Adaptive Fault Detection and Classification

• ConclusionsConclusions

Outline

©2010 Mladen Kezunovic, All Rights Reserved2

• Data IntegrationI f i E h• Information Exchange

• Temporal ConsiderationsS ti l C id ti• Spatial Considerations

Background


Data Integrationg

CFL MS PE EMSLEVEL ICENTRALIZED

RC

CENTRALIZED LOCATION

LMS

IS SCLEVEL IISUBSTATION

FL DFR CBM DPR RTU SOELEVEL III FL DFR CBM DPR RTU SOE

A S A S A AS AS AS

SWITCHYARD INTERFACE


Information Exchangeg

Database:Raw and pre‐processed measurementsSystem configuration data

5©2010 Mladen Kezunovic, All Rights Reserved

Information Exchange

i h

g

Engineer Dispatcher Technician

SummaryReport

ComprehensiveReport

Brief Report

Reportp


Temporal Considerations


Temporal Considerations

• Relative and absolute time as a reference for correlating power system events

• Sampling clock time as a reference for synchronous signal• Sampling clock time as a reference for synchronous signal sampling vs. scanning

• Time window as a reference for waveform representation in time and frequency domain

• Implementation of the time reference


Synchronous Sampling vs. Scanning


GPS Synchronization


Spatial Considerations

Network





• Location as a reference for data processing and information extraction

• Location as a reference for model representation• Location as a reference for model representation

Application Temporal Spatial Model

Optimal Fault Location

Synchronized or unsynchronized phasor or sample

t

Local and system‐wide

Power System Network for short circuit study

vectory

Intelligent Alarm Processing

Synchronized or unsynchronized

Substation and system‐wide

Petri‐Net Logic for cause‐effect

Processingphasors

system widerepresentation

Inherently Adaptive Fault Synchronized

LocalPower system model for training


Detection and Classification

sample vectorLocal model for training

pattern clustering

• OverviewT l i l k l d• Translational knowledge− Data Processing− Information ExtractionInformation Extraction− Implementation

• Case Studyy

Optimal Fault Location


Overview of Fault Location Algorithms

• Phasor based Methods• Phasor based MethodsUse fundamental frequency component of the signal and lumped parameter model

o Single endo Double end

• Time-domain based MethodsUse transient components of the signal and l d di ib d d l

o Synchronized

o Double end

lumped or distributed parameter model• Traveling wave based Methods

Use correlation between the forward and

o Unsynchronized

o PhasorsUse correlation between the forward and backward travelling waves along a line or direct detection of the arrival time

o Samples


An Optimal Solution

• Substation Data Flow• Substation Data Flow


Data Processing

• System level data• System level data− Power system model data− PI Historian datasto a data− Sequence data

• Field data (recorded by IEDs)− Event data− Waveform data

• Substation interpretation data


Information Extraction

• Extraction of phasors• Extraction of phasors− Remove high-frequency noise− Remove decaying dc-offsete ove decay g dc o set

• Synchronization of phasors• Tuning the power system model with real-time power system

conditions− Updating power grid topology

U d ti ti d l d d t− Updating generation and load data


Translational Knowledge through matching system wide sparse measurement with power system model


Implementation


Case Study

Evaluate the following issues:Evaluate the following issues:

• Various number of files• Specifying the search region• Using preprocessed fault location estimation• Using different quantities for the match• Evaluating differences in the accuracyOne test case provided by the utility:

• Two subsequent (in 5 ms gap) phase to ground faultsTwo subsequent (in 5 ms gap) phase to ground faults• PI Historian data provided for two substations• DFR triggered for only one substation



• Case Studyy

Intelligent Alarm Processing


Overview of Alarm ProcessingIssues that operators face:

Al hi h t d i ti h• Alarms which are not descriptive enough• Alarms which are too detailed• Too many alarms during a system disturbance• False alarms• Multiplicity of alarms for the same event• Alarms changing too fast to be read on the displayAlarms changing too fast to be read on the display• Alarms not in priority orderAlarm processing algorithms:

• Expert System (ES) technique• Expert System (ES) technique• Fuzzy Logic (FL) technique• Petri-Nets (PN) technique


• Fuzzy Reasoning Petri-Nets (FRPN) technique

Solution by Matching Data with Models

This approach assumes that filed data and model of the relay actions are matchedThis approach assumes that filed data and model of the relay actions are matched leading to a cause‐effect analysis.

• Suppress multiple alarms from one event;G t i l l i th h l i l ff t• Generate a single conclusion through logical cause-effect relationship;

• Automate the process to get answers quickly;p g q y;• Make graphical and numerical information concise and easy to

follow.


Data Processing• Input data list:

Data from RTU of SCADA (Main data)

1 CB status change alarms (Opening and Closing)

2 Trip signal of main transmission line relays2 Trip signal of main transmission line relays

3 Trip signal of primary backup transmission line relays

4 Trip signal of secondary backup transmission line relays

5 Trip signal of bus relays

Data from Digital Protective Relays (Additional data)

1 Pickup & Operation signals of main transmission line relays1 Pickup & Operation signals of main transmission line relays

2 Pickup & Operation signals of primary backup transmission line relays

3 Pickup & Operation signals of secondary backup transmission line relays

4 Pi k & O i i l f b l


4 Pickup & Operation signals of bus relays


• Stage I: System topology is analyzed based on CBs status data;

• Stage II: FRPN diagnosis and operation.


ImplementationTranslational Knowledge through matching SCADA and relay data with theTranslational Knowledge through matching SCADA and relay data with the reasoning diagnosis model


Implementation

p1(SLR4210 Trip)p1(SLR4210 Trip)

p2(CB4210 Open) r1 p13

1121

61

131

21c

p6(CB4160 Open)6

13 1 11 2 21 6 61 1[ (1 ) ]c

• A FRPN model for BBSES_60A fault • An Example of “Weighted Average” Operation


Case Study

• Alarm Screen Shot


Case 1:

Condition: No protective relay signals CB4210 4220 4160 4920 status changesCondition: No protective relay signals. CB4210, 4220, 4160 4920 status changes are detected.Diagnosis result: Line BBSES_60A is faulted, and its truth value is 0.5130.


Case 2:Condition: Operation of CB is tripped by associated relays. Received relays signals. CB4210 4220 4160 4920 t t h d t t dCB4210, 4220, 4160 4920 status changes are detected.Diagnosis result: Line BBSES_60A is faulted, and its truth value is 0.8550. with the input of related relay signals, the fault certainness has been increased dramatically.


Advantages

• The fault alarm analysis report can be generated automatically• The fault alarm analysis report can be generated automatically and immediately after the fault occurs.

• The FRPN models can be built in advance based on power psystem and protection system configurations and stored in files.

• The FRPN models can be easily modified according to the h f i t d t ll t d t tichanges of input data as well as power system and protection

system configuration.• This solution can use only SCADA data and does not need y

detailed data from IEDs or other measurement devices.



• Case Study

Inherently Adaptive Fault

y

Inherently Adaptive FaultDetection and Classification


Overview of Neural Network Algorithms


Fuzzy ART Neural Network Algorithm

D t F P Fuzzy ARTData From Power System Simulation

Data From Substation Historical

ART Neural Network Training

Fuzzy ART Neural Network

Algorithm

Databaseg

Prototypes ofPrototypes of trained clusters

Data From RealSystem , CVT, CT

Selection of K nearest neighbour

Clusters

Fuzzy K -NN Classification

Fault Detection,Type & Zone


Data Processing




Implementation

Detecting and Classifying Faults by Mapping Data to Labeled ClustersDetecting and Classifying Faults by Mapping Data to Labeled Clusters

• The raw training patterns ‐ information • The patterns are allocated to the clusters after a training processing ‐ knowledge


Fuzzification of NN outputs


Final Implementation


Case Study

Error results of neural network fault classification toolsError results of neural network fault classification tools


• Fault location: matching field data with model data creates an

ti l l tioptimal solution ;• Alarm processing: matching

data to the model of relaying logic creates an intelligent cause-effect analysis;

• Protective relaying: matching y g gdata vectors from the field signals with cluster of patterns designating fault types created an

Conclusions

g g ypinherently adaptive protection.


Thank you!Questions?

Mladen KezunovicTel: (979) 845‐7509E‐mail: [email protected]


Documents

Translational Knowledge: From Collecting Data to Making